But specifically, for the "quadriplegic" tag, I really wanted to get her legs to be near the chair legs and not in some other pose. There is just not a lot of good data of women sitting in wheel chairs.
The intent of trying to get this to work is to have some bondage out in public locations, but it is difficult to force it.
As an interesting note, quadriplegic led to a significant increase in how often the image ended up being in black and white or just very low contrast. The word embeddings for it have a ton of connotations with it and no representation within Booru tags, so I stopped using the tag.
This is from a giant script that uses sd-dynamic-prompts in Automatic1111 to generate different scenarios. So there's different professions, hair styles, bondage situations, etc.
If I am going for a specific scenario, it is easy to clean up to get something readable.
This is one of the best pieces I've see here. I've been trying to replicate this style of face to no avail. I've used the same model version, the newer one, and a combo of the loras indicated below. Any tips?
@abi_00 Sorry, I missed your comment earlier. Could be a number of things really. I tried the same settings on v12 of realcartoon and it was very similar so it's unlikely that. Maybe it's clip skip or missing embeddings (BadDream is the most important one for the style I think)? Could also be that some of the loras I'm using have new versions and they look different. I think I didn't do anything weird during upscaling so that's not a problem, but maybe you're starting with a different resolution and that makes the faces different. You could always send me the image you're getting, here or on discord and we can troubleshoot from there.
